Where should an Australian SME start with AI?

Begin where value is getting stuck

Most businesses first encounter AI as a catalogue of tools. That encourages the wrong opening question: ‘What could we use AI for?’ A better question is: ‘Where is the business repeatedly losing time, margin, customer momentum or decision quality?’

Good starting points are visible and consequential. Quotes take too long to become jobs. Customer requests lose context between the inbox and the project system. Senior people spend hours reconstructing information before a decision. Work is entered twice because two systems do not agree. These are business constraints first; AI is only one possible response.

Build a baseline before proposing a solution

A useful baseline does not need to be perfect. It needs enough evidence to show the current volume, delay, manual effort, error rate or commercial leakage. Pull the signal from the systems that already record the work and test it against what the people doing the work know.

The mismatch between the system and the lived experience is often the most valuable finding. It may reveal poor data capture, an unofficial workaround or a handoff that exists in practice but not in the documented process.

• Name the decision or workflow that needs to improve.
• Identify the systems and people that hold evidence about it.
• Measure the current state using a small number of meaningful indicators.
• Test the suspected cause before choosing technology.

Choose the smallest change that can prove value

Once the cause and value are clearer, test options in order: improve the process, configure technology you already own, buy a proven product, customise a repeatable component, then consider a fully custom build. Custom AI should earn its place rather than being the default.

The first implementation should be narrow enough to measure and important enough to matter. Define what success looks like, where a person remains responsible, how exceptions are handled and what would make you reverse the change.

What a sound first engagement produces

The output should be more than a list of AI ideas. It should leave the business with a connected view of the evidence, a map of the workflow, a ranked set of opportunities and one practical first move. If the business proceeds, the implementation should be documented and handed over so the capability can be operated without permanent dependency on the agency that built it.